Algorithms For Wireless Sensor Networks

Sartaj Sahni and Xiaochun Xu Department of Computer and Information Science and Engineering, University of Florida, Gainesville, FL 32611 {sahni,xxu}@cise.u.edu September 7, 2004
Abstract This paper reviews some of the recent advances in the development of algorithms for wireless sensor networks. We focus on sensor deployment and coverage, routing and sensor fusion.

Keywords:

Wireless sensor networks, algorithms, routing, coverage, fusion.

Introduction

A wireless sensor network may comprise thousands of sensor nodes. Each sensor node has a sensing capability as well as limited energy supply, compute power, memory and communication ability. Besides military applications, wireless sensor networks may be used to monitor microclimates and wildlife habitats [56], the structural integrity of bridges and buildings, building security, location of valuable assets (via sensors placed on these valuable assets), trac, and so on. However, realizing the full potential of wireless sensor networks poses myriad research challenges ranging from hardware and architectural issues, to programming languages and operating systems for sensor networks, to security concerns, to algorithms for sensor network deployment, operation and management. Iyengar and Brooks [26, 27] and Culler and Hong [12] provide good overviews of the breadth of sensor network research topics as well as of applications for sensor networks. This paper focuses on some of the algorithmic issues that arise in the context of wireless sensor networks. Specically, we review algorithmic issues in sensor deployment and coverage, routing, and fusion. There is an abundance of algorithmic research related to wireless sensor networks. At a high level, the developed algorithms may be categorized as either centralized or distributed. Because of the limited memory, compute and communication capability of sensors, distributed algorithms research has focused on localized distributed algorithmsdistributed algorithms that require only local (e.g., nearest neighbor) information. 1

Sensor Deployment and Coverage

In a typical sensor network application, sensors are to be placed (or deployed) so as to monitor a region or a set of points. In some applications we may be able to select the sites where sensors are placed while in others (e.g., in hostile environments) we may simply scatter (e.g., air drop) a suciently large number of sensors over the monitoring region with the expectation that the sensors that survive the air drop will be able to adequately monitor the target region. When site selection is possible, we use deterministic sensor deployment and when site selection isnt possible, the deployment is nondeterministic. In both cases, it often is desirable that the deployed collection of sensors be able to communicate with one another, either directly or indirectly via multihop communication. So, in addition to covering the region or set of points to be sensed, we often require the deployed collection of sensors to form a connected network. For a given placement of sensors, it is easy to check whether the collection covers the target region or point set and also whether the collection is connected. For the coverage property, we need to know the sensing range of individual sensors (we assume that a sensor can sense events that occur within a distance r , where r is the sensors sensing range, from it) and for the connected property, we need to know the communication range, c, of a sensor. Zhang and Lou [72] have established the following necessary and sucient condition for coverage to imply connectivity. Theorem 1 [Zhang and Lou [72]] When the sensor density (i.e., number of sensors per unit area) is nite, c 2r is a necessary and sucient condition for coverage to imply connectivity. Wang et al. [61] prove a similar result for the case of k-coverage (each point is covered by at least k sensors) and k-connectivity (the communication graph for the deployed sensors is k connected). Theorem 2 [Wang et al. [61]] When c 2r , k-coverage of a convex region implies k-connectivity. Notice that k-coverage with k > 1 aords some degree of fault tolerance, we are able to monitor all points so long as no more than k 1 sensors fail. Huang and Tseng [25] develop algorithms to verify whether a sensor deployment provides k-coverage. Other variations of the sensor deployment problem also are possible. For example, we may have no need for sensors to communicate with one another. Instead, each sensor communicates directly with a base station that is situated within the communication range of all sensors. In another variant [23, 24], the sensors are mobile and self deploy. A collection of mobile sensors may be placed into an unknown and potentially hazardous environment. Following this initial placement, the sensors relocate so as to obtain maximum coverage of the unknown environment. They 2

Step 1: [Achieve Coverage] 3 Let = ( 2 + 1)r . Place a sensor at (i, j), i even and j integer as well as one at (i + r/2, j), i odd and j integer. Step 2: [Achieve Connectivity] 3 Let = 2 r . Place a sensor at (0, j ), j odd. Figure 1: Kar and Banerjees sensor-deployment algorithm communicate the information they gather to a base station outside of the environment being sensed. A distributed potential-eld-based algorithm to self deploy mobile sensors under the stated assumptions is developed in [24] and a greedy and incremental self-deployment algorithm is developed in [23]. A virtual-force algorithm to redeploy sensors so as to maximize coverage also is developed by Zou and Chakrabarty [73]. Poduri and Sukhatme [47] develop a distributed self-deployment algorithm that is based on articial potential elds and which maximizes coverage while ensuring that each sensor has at least k other sensors within its communication range.

2.1
2.1.1

Deterministic Deployment
Region Coverage

Kar and Banerjee [32] examine the problem of deploying the fewest number of homogeneous sensors so as to cover the plane with a connected sensor network. They assume that the sensing range equals the communication range (i.e., r = c). Figure 1 gives their deployment algorithm. One may verify that the sensors deployed in Step 1 are able to sense the entire plane. So, these sensors satisfy the coverage requirement. However, the sensors placed in Step 1 dene many rows of connected sensors with the property that two sensors in dierent rows are unable to communicate (i.e., there is no multihop path between the sensors such that adjacent sensors on this path are at most c apart). Step 2 creates a connected network by placing a column of sensors in such a way as to connect together the connected rows that result from Step 1. Kar and Banerjee [32] have shown that their algorithm of Figure 1 has a sensor density that is within 2.6% of the optimal density. This algorithm may be extended to provide connected coverage for a set of nite regions [32].

Step 1: [Initialize] Let s be any leaf of the Euclidean minimum-cost spanning tree of the point set. candidateSet = {s} Step 2: [Deploy Sensors] while (candidateSet = ) { Remove any point p from candidateSet. Place a sensor at p. Remove from candidateSet all points covered by the sensor at p. Add to candidateSet all points (not necessarily vertices) q on the spanning tree T that satisfy the conditions: (1) q is distance r from p. (2) q is not covered by an already placed sensor. (3) The spanning tree path from s to q is completely covered by already placed sensors. }

Figure 2: Greedy algorithm of [32] to deploy sensors 2.1.2 Point Coverage

Figure 2 gives the greedy algorithm of Kar and Banerjee [32] to deploy a connected sensor network so as to cover a set of points in Euclidean space. This algorithm, which assumes that r = c, uses at most 7.256 times the minimum number of sensors needed to cover the given point set [32]. It is easy to see that the constructed deployment covers all of the given points and is a connected network. Grid coverage is another version of the point coverage problem. In this version, Chakrabarty et al. [7], we are given a two- or three-dimensional grid of points that are to be sensed. Sensor locations are restricted to these grid points and each grid point is to be covered by at least m, m 1, sensors (i.e., we seek m-coverage). For sensing, we have k sensor types available. A sensor of type i costs ci dollars and has a sensing range ri . At most one sensor may be placed at a grid point. In this version of the point coverage problem, the sensors do not communicate with one another and are assumed to have a communication range large enough to reach the base station from any grid position. So, network connectivity is not an issue. The objective is to nd a least-cost sensor deployment that provides m-coverage. Chakrabarty et al. [7] formulate this m-coverage deployment problem as an integer linear program (ILP) with O(kn2 ) variables and O(kn2 ) equations, where n is the number of grid points. Xu and Sahni [66] reduce the number of variables to O(kn) and the number of equations to O(n). Also, their formulation doesnt require the sensor locations and points to be sensed to form a grid. Let sij be a

0/1-valued variable with the interpretation sij = 1 i a sensor of type i is placed at point j , 1 i k, 1 j n. The solution to the following ILP describes an optimal sensor deployment.
k n

minimize
i=1 k

(ci
j =1

sij )

sia m, 1 j n
i=1 aX (i,j ) k

sij 1, 1 j n
i=1

where X (i, j ) = {all points within ri of point j } Even with this reduction in the number of variables and equations, the ILP is practically solvable only for a small number of points n. For large n, Chakrabarty et al. [7] propose a divide-and-conquer near-optimal algorithm in which the base case (small number of points) is solved optimally using the ILP formulation.

2.2

Maximizing Coverage Lifetime

When sensors are deployed in dicult-to-access environments, as is the case in many military applications, a large number of sensors may be air-dropped into the region that is to be sensed. Assume that the sensors that survive the air drop cover all targets that are to be sensed. Since the power supply of a sensor cannot be replenished, a sensor becomes inoperable once it runs out of energy. Dene the life of a sensor network to be the earliest time at which the network ceases to cover all targets. The life of a network can be increased if it is possible to put redundant sensors (i.e., sensors not needed to provide coverage of all targets) to sleep and awaken these sleeping sensors when they are needed to restore target coverage. Sleeping sensors are inactive while sensors that are awake are active. Inactive sensors consume far less energy than active ones. Cardei and Du [5] propose partitioning the set of available sensors into disjoint sets such that each set covers all targets. Let T be the set of targets to be monitored and let Si denote the subset of T in the range of sensor i, 1 i n. Let P1 , P2 , , Pk be disjoint partitions of the set of n sensors such that j Pi Sj = T , 1 i k. Then the set of sensors in each Pi covers all targets. We refer to the set of Pi s as a disjoint set cover of size k. Moreover, by going through k sleep/awake rounds where in round i only the sensors in Pi are awake, we are able to monitor all targets in each round and increase the network life 5

to almost kt, where t is the time it takes a sensor to deplete its energy when in the awake mode. Since sensors deplete energy even when in the sleep mode, the life of each round is slightly less than that of the preceding round. Cardei and Du [5] have shown that deciding whether there is a disjoint set cover of size k for a given sensor set is NP-complete. They develop a heuristic to maximize the size of a disjoint set cover. An experimental evaluation of this heuristic reveals that it nds about 10% more disjoint covers than does the best algorithm of Slijepcevic and Potkonjak [53]. However, the algorithm of Cardei and Du [5] takes more time to execute. Several decentralized localized protocols to control the sleep/awake state of sensors so as to increase network lifetime have been proposed. Ye, Zhong, Lu and Zhang [70] propose a very simple protocol. In this protocol, the set of active nodes provide the desired coverage. A sleeping node wakes up when its sleep timer expires and broadcasts a probing signal a distance d (d is called the probing range). If no active sensor is detected in this probing range, the sensor moves into the active state. However, if an active sensor is detected in the probing range, the sensor determines how long to sleep, sets its sleep timer and goes to sleep. Techniques to dynamically control the sleep time and probing range are discussed in [70]. Simulations reported in [70] indicate that this simple protocol outperforms the GAF protocol of [67]. However, experiments conducted by Tian and Georganas [57] reveal that the protocol of Ye et al. [70] cannot ensure the original sensing coverage and blind spots may appear after turning o some nodes. Tian and Georganas [57] propose an alternative distributed localized protocol that sensors may use to turn themselves on and o. The network operates in rounds, where each round has two phasesself-scheduling and sensing. In the self-scheduling phase each sensor decides whether or not to go to sleep. In the sensing phase, the active/awake sensors monitor the region. Sensor s turns itself o in the self-scheduling phase if its neighbors are able to monitor the entire sensing region of s. To make this determination, every sensor broadcasts its location and sensing range. A backo scheme is proposed to avoid blind spots that would otherwise occur if two sensors turn o simultaneously, each expecting the other to monitor part or all of its sensing region. In this backo scheme, each active sensor uses a random delay before deciding whether or not it can go to sleep without aecting sensing coverage. The decentralized algorithm OGDC (Optimal Geographical Density Control) of Zhang and Lou [72] guarantees coverage, which by Theorem 1 implies connectivity whenever the sensor communication range is at least twice its sensing range. Experimental results reported in [72] suggest that when OGDC is used, the number of active (awake) nodes may be up to half that when the PEAS [69] or GAF [67] algorithms are used to control sensor state. The Coverage Conguration Protocol (CCP) to maximize

lifetime while providing k-coverage (as well as k-connectivity when c 2r , Theorem 2) is developed in [61]. A distributed protocol for dierentiated surveillance is proposed by Yan, He and Stankovic [68]. Assume that the probability that sensor s detects an event at a distance d is P (s, x) = 1/(1 + d) . The probability P (x) that an event at x is detected by the sensor network is P (x) = 1 (1 P (s, x)) where the product is taken over all sensors in the network. The coverage, C , of the sensor network (overall network coverage) is the sum of P (x) over all points in the sensing region. Let C (s) be the overall coverage when sensor s is removed from the network. The sensing denomination [37] of a sensor s is its contribution, C C (s), to overall network coverage. Lu and Suda [37] use the sensing denomination of a sensor to obtain a distributed localized self-scheduling algorithm to schedule the sleep/awake states of sensors so as to increase network lifetime. In their algorithm, each sensor periodically makes a decision to either go to sleep or be active. Sensors with higher sensing denomination have a higher probability of being active.

2.3

Deployment Quality

The quality of a sensor deployment may be measured by the minimum k for which the deployment provides k-coverage as well as by the minimum k for which we have k-connectivity. By Theorem 2 the rst metric implies the second when the communication range c is at lest twice the sensing range r . Meguerdichian et al. [42] have formulated additional metrics suitable for a variety of sensor applications. In these applications, the ability of a sensor to detect an activity a distance d from the sensor is given by the function /(1 + d) , where and are constants. So, sensing ability is maximum when d = 0 and declines as we get farther from the sensor. Let P be a path that connects two points u and v (these points may be within or outside the region being sensed). The breach weight, BW (P, u, v ), of P is the closest path P gets to any of the deployed sensors. The breach weight, BW (u, v ), of the points u and v is the maximum of the breach weights of all paths between u and v . BW (u, v ) = max{BW (P, u, v )|P is a path between u and v } The breach weight or breachability, BW, of a sensor network is the maximum of the breach weights of all point pairs. BW = min{BW (u, v )| u and v are points on the boundary of the sensing region} 7

When the ability of a sensor to detect an activity is inversely proportional to some power k of distance, sensor deployments that minimize breach weight are preferred. The breachability of a network gives us an indication of how successful an intruder could be in evading detection. Suppose we have an application in which items are being transported between pairs of points and our sensors track the progress of these shipments. Now we wish to use maximally observable paths, that is paths that remain as close to a sensor as possible. Let d(x) be the distance between a point x and the sensor nearest to x. Meguerdichian et al. [42] dene the support weight, SW (P, u, v ), of a path P between u and v as

SW (P, u, v ) = max{d(x)|x is a point on P } SW (u, v ) now is dened as SW (u, v ) = min{SW (P, u, v )|P is a path between u and v } and the support weight (or simply support), SW , of the sensor network is SW = max{SW (u, v )| u and v are points on the boundary of the sensing region} Although Meguerdichian et al. [42] develop centralized algorithms to compute BW (u, v ) and SW (u, v ), these algorithms are awed [35]. Li, Wan and Frieder [35] describe a distributed localized algorithm to determine SW (u, v ). This algorithm is given in Figure 3. In this algorithm |sa| is the Euclidean distance between the sensors s and a. The algorithm assumes that u and v are in the convex hull of the sensor locations. Since there may be several paths with the computed SW (u, v ) value, we may be interested in nding, say, a best support path from u to v that has minimal length. Li et al. [35] develop a localized distributed algorithm to nd an approximately minimal length path with support SW (u, v ). The exposure E (P, u, v ) of a path P from u to v is dened as
v

E (P, u, v ) =
u

S (x)dx

where S (x) is a sensing function and the integral is computed over the path P . Meguerdichian et al. [43] suggest two sensing functions. One is simply the sensing ability of the closest sensor to x; the other is the sum of the abilities of all sensors to detect activity at x. An intruder who wishes to minimize the risk of detection would take a minimal exposure path to get from u to v . Figure 4 gives the algorithm of [43] to nd an approximation to the minimal-exposure path between u and v . This algorithm overlays 8

Step 1: [Construct Local Neighborhood Graph] Each sensor broadcasts its id and location. Each sensor s compiles a list L(s) of all ids and locations that it hears. Let A(s), the adjacency list for s, comprise all sensors a L(s) such that there is no b L(s) located in the interior of the intersection region of the radius |sa| circles centered at s and a. For each a A(s), the weight of the edge (s, a) is |sa|/2. Step 2: [Construct Best Support Path] Let the length of a path be the maximum weight of its edges. Let x and y , respectively, be the sensors closest to the points u to v . Run the distributed Bellman-Ford shortest path algorithm to determine a shortest path P (x, y ), in the local neighborhood graph, from x to y . (u, x), P (x, y ), (y, v ) is a best support path from u to v . The weight of (u, x) is |ux| and that of (y, v ) is |yv |. SW (u, v ) is the maximum of the edges weights in the best support path. Figure 3: Distributed algorithm of [35] to compute SW (u, v ) Step 1: [Graph Overlay] Overlay the sensing region with a weighted undirected graph G. The vertices of G are points in the sensing region and its edges are straight lines. The weight of an edge is the exposure of that edge. Step 2: [Minimal-Exposure Path] The minimal-exposure path from u to v is estimated to be the shortest path in G from the vertex of G closest to the point u to the vertex of G closest to the point v . Its exposure is the length of this shortest path. Figure 4: Algorithm of [43] to nd an approximate minimal-exposure path an undirected weighted graph over the sensing region and then nds a shortest path from the graph vertex nearest to u to the graph vertex nearest to v . This shortest path may be found using Dijkstras shortest path algorithm. By increasing the number of vertices and edges used in Step 1, the accuracy of the approximation is increased. For the overlay graph, [43] considers a uniform two-dimensional grid arrangement of the vertices and a variety of options for edges (e.g., edges connect a vertex to its north, south, east, west, northwest, northeast, southeast and southwest neighbors). Sensor deployments that maximize minimal-exposure are preferred. Veltri et al. [59] develop a distributed localized algorithm to nd an approximate minimal-exposure path. Also, they show that nding a maximal-exposure path is NP-hard and they propose several heuristics to construct approximate maximal-exposure paths. Additionally, a linear programming formulation

for minimal- and maximal-exposure paths is obtained. Kanan et al. [29] develop polynomial time algorithms to compute the maximum vulnerability of a sensor deployment to attack by an intelligent adversary and use this to compute optimal deployments with minimal vulnerability.

Routing

Traditional routing algorithms for sensor networks are datacentric in nature. Given the unattended and untethered nature of sensor networks, datacentric routing must be collaborative as well as energyconserving for individual sensors. Kannan et al. [28, 30] have developed a novel sensor-centric paradigm for network routing using game-theory. In this sensor-centric paradigm, the sensors collaborate to achieve common network-wide goals such as route reliability and path length while minimizing individual costs. The sensor-centric model can be used to dene the quality of routing paths in the network (also called path weakness). Kannan et al. [30] describe inapproximability results on obtaining paths with bounded weakness along with some heuristics for obtaining strong paths. The development of ecient distributed algorithms for approximately optimal strong routing is an open issue that can be explored further. Energy conservation is an overriding concern in the development of any routing algorithm for wireless sensor networks. This is because such networks are often located such that it is dicult, if not impossible, to replenish the energy supply of a sensor. Three formsunicast, broadcast and multicastof the routing problem have received signicant attention in the literature. The overall objective of these algorithms is to either maximize the lifetime (earliest time at which a communication fails) or the capacity of the network (amount of data trac carried by the network over some xed period of time). Assume that the wireless network is represented as a weighted directed graph G that has n vertices/nodes and e edges. Each node of G represents a node of the wireless network. The weight w(i, j ) of the directed edge (i, j ) is the amount of energy needed by node i to transmit a unit message to node j. In the most common model used for power attenuation, signal power attenuates at the rate a/r d , where a is a media dependent constant, r is the distance from the signal source, and d is another constant between 2 and 4 [48]. So, for this model, w(i, j ) = w(j, i) = c r (i, j )d , where r (i, j ) is the Euclidean distance between nodes i and j and c is a constant. In practice, however, this nice relationship between w(i, j ) and r (i, j ) may not apply. This may, for example, be due to obstructions between the nodes that may cause the attenuation to be larger than predicted. Also, the transmission properties of the media may be asymmetric resulting in w(i, j ) = w(j, i).

10

3.1

Unicast

In a unicast, we wish to send a message from a source sensor s to a destination sensor t. Singh, Woo and Raghavendra [52] propose ve strategies that may be used in the selection of the routing path for this transmission. The rst of these is to use a minimum-energy path (i.e., a path in G for which the sum of the edge weights is minimum) from s to t. Such a path may be computed using Dijkstras shortest path algorithm [49]. However, since, in practice, messages between several pairs of source-destination sensors need to be routed in succession, using a minimum-energy path for a message may prevent the successful routing of future messages. As an example, consider the graph of Figure 5. Suppose that sensors x, b1 , , bn initially have 10 units of energy each and that u1 , , un each have 1 unit. Assume that the rst unicast is a unit-length message from x to y . There are exactly two paths from x to y in the sensor network of Figure 5. The upper path, which begins at x, goes through each of the ui s, and ends at y uses n + 1 energy units; the lower path uses 2(n + 1) energy units. Using the minimum energy path, depletes the energy in node ui , 1 i n. Following the unicast, sensors u1 , , un are unable to forward any messages. So an ensuing request to unicast from ui to uj , i < j will fail. On the other hand, had we used the lower path, which is not a minimum energy path, we would not deplete the energy in any sensor and all unit-length unicasts that could be done in the initial network also can be done in the network following the rst x to y unicast.

1 x 2

u1

u2

...

un

1 2 y

b1

b2

...

bn

Figure 5: A sensor network

The remaining four strategies proposed in [52] attempt to overcome the myopic nature of the minimumenergy-path strategy, which sacrices network lifetime and capacity in favor of total remaining energy. Since routing decisions must be made in an online fashion (i.e., if the ith message is to be sent from si to ti , the path for message i must be decided without knowledge of sj and tj , j > i), we seek an online algorithm with good competitive ratio. It is easy to see that there can be no online unicast algorithm with constant competitive ratio with respect to network lifetime and capacity [1]. For example, consider

11

the network of Figure 5. Assume that the energy in each node is 1 unit. Suppose that the rst unicast is from x to y . Without knowledge of the remaining unicasts, we must select either the upper or lower path from x to y . If the upper path is chosen and the source-destination pairs for the remaining unicasts turn out to be (u1 , u2 ), (u2 , u3 ), , (un1 , un ), (un , y ) then the online algorithm routes only the rst unicast whereas an optimal oine algorithm would route all n + 1 unicasts, giving a competitive ratio of n + 1. The same ratio results when the lower path is chosen and the source-destination pairs for the remaining unicasts are (b1 , b2 ), (b2 , b3 ), , (bn1 , bn ), (bn , y ). Theorem 3 There is no online algorithm to maximize lifetime or capacity that has a competitive ratio smaller than (n). To maximize lifetime and/or capacity, we need to achieve some balance between the energy consumed by a route and the minimum residual energy at the nodes along the chosen route. Aslam, Li and Rus [1] propose the max-min zPmin -path algorithm to select unicast routes that attempt to make this balance. This algorithm selects a unicast path that uses at most z Pmin energy, where z is a parameter to the algorithm and Pmin is the energy required by the minimum-energy unicast path. The selected unicast path maximizes the minimum residual energy fraction (energy remaining after unicast/initial energy) for nodes on the unicast path. Notice that the possible values for the residual energy fraction of node u may be obtained by computing (ce(u) w(u, v ) l)/ie(u), where l is the message length, ce(u) is the (current) energy at node u just before the unicast, ie(u) is the initial energy at u, and w(u, v ) is the energy needed to send a unit-length message along the edge (u, v ). This computation is done for all vertices v adjacent from u. Hence the union, L, of these values taken over all u gives the possible values for the minimum residual-energy fraction along any unicast path. Figure 6 gives the max-min zPmin algorithm. Several adaptations to the basic max-min zPmin algorithm, including a distributed version are described in [1]. Kar et al. [33] develop an online capacity-competitive algorithm, CMAX, with logarithmic competitive ratio. On the surface, this would appear to violate Theorem 3. However, to achieve this logarithmic competitive ratio, the algorithm CMAX does admission control. That is, it rejects some unicasts that are possible. The bound of Theorem 3 applies only for online algorithms that must perform every unicast that is possible. Let ce, E and l be as for the max-min zPmin algorithm. Let (u) = 1 ce(u)/ie(u) be the fraction of us initial energy that has been used so far. Let and be two constants. In the CMAX algorithm, the weight of every edge (u, v ) is changed from w(u, v ) to w(u, v ) ((u) 1). The shortest source-todestination path P in the resulting graph is determined. If the length of this path is more than , the 12

Step 1: [Initialize] Eliminate from G every edge (u, v ) for which ce(u) < w(u, v ) l. Let L be the list of possible values for the minimum residual-energy fraction. Step 2: [Binary Search] Do a binary search in L to nd the maximum value max of the minimum residual-energy fraction for which there is a path P from source to destination that uses at most z Pmin energy. For this, when testing a value q from L, we nd a shortest source to destination path that does not use edges (u, v ) that make the residual-energy fraction at u less than q . Step 3: [Wrap Up] If no path is found in Step 2, the unicast isnt possible. Otherwise, use the path P corresponding to max. Figure 6: The max-min zPmin unicast algorithm of [17] Step 1: [Initialize] Eliminate from G every edge (u, v ) for which ce(u) < w(u, v ) l. Change the weight of every remaining edge (u, v ) to w(u, v ) ((u) 1). Step 2: [Shortest Path] Let P be the shortest source-to-destination path in the modied graph. Step 3: [Wrap Up] If no path is found in Step 2, the unicast isnt possible. If the the length of P is more than do not do the unicast. Otherwise, use P for the unicast. Figure 7: CMAX algorithm of [33] for unicasts unicast is rejected (admission control); otherwise, the unicast is done using path P . Figure 7 gives the algorithm. The CMAX algorithm of Figure 7 has a complexity advantage over the max-min zPmin algorithm of Figure 6. The former does only 1 shortest-path computation per unicast while the latter does O(logn), where n is the number of sensor nodes. Although admission control is necessary to establish the logarithmic competitive-ratio bound for CMAX, we may use CMAX without admission control (i.e., route very unicast that is feasible) by setting = . Experimental results reported in [33] suggest that CMAX with no admission control outperforms max-min zPmin on both the lifetime and capacity metrics. In the MRPC lifetime-maximization algorithm of Misra and Banerjee [45], the capacity, c(u, v ) of 13

Step 1: [Initialize] Eliminate from G every edge (u, v ) for which ce(u) < w(u, v ) l. For every remaining edge (u, v ) let c(u, v ) = ce(u)/w(u, v ). Let L be the list of distinct c(u, v ) values. Step 2: [Binary Search] Do a binary search in L to nd the maximum value max for which there is a path P from source to destination that uses no edge with c(u, v ) < max. For this, when testing a value q from L, we perform a depth- or breadth-rst search beginning at the source. The search is not permitted to use edges with c(u, v ) < q . Let P be the source-to-destination path with lifetime max. Step 3: [Wrap Up] If no path is found in Step 2, the unicast isnt possible. Otherwise, use P for the unicast. Figure 8: MRPC algorithm of [45] for unicasts edge (u, v ) is dened to be ce(u)/w(u, v ). Note that c(u, v ) is the number of unit-length messages that may be transmitted along (u, v ) before node u runs out of energy. The lifetime of path P , lif e(P ) is dened to be the the minimum edge capacity on the path. In MRPC, the unicast is done along a path P with maximum lifetime. Figure 8 gives the MRPC unicast algorithm. A decentralized implementation as well as a conditional MRPC in which minimum-energy routing is used so long as the selected path has a lifetime that is greater than or equal to a specied threshold. When the lifetime of the selected path falls below this threshold, we switch to MRPC routing. Chang and Tassiulas [8, 9] develop a linear-programming formulation for lifetime maximization. This formulation requires knowledge of the rate at which each node generates unicast messages. Wu, Gao and Stojmenovic [65] propose unicast routing based on connected dominating sets to maximize network lifetime. Stojmenovic and Lin [55] and Melodia et al. [44] develop localized unicast algorithms to maximize lifetime and Heinzelman, Chandrakasan and Balakrishnan [22] develop a clustering-based routing algorithm (LEACH) for sensor networks..

3.2

Multicast and Broadcast

Using an omnidirectional antenna, node i can transmit the same unit message to nodes j1 , j2 , , jk , using ewireless = max{w(i, jq )|1 q k}

14

energy rather than

k

ewired =
q =1

w(i, jq )

energy. Since, ewireless ewired , the reduction in energy needed to broadcast from one node to several others in a wireless network over that needed in a wired network is referred to as the wireless broadcast advantage [62, 63]. Because of the similarity between multicast and broadcast algorithms for wireless sensor networks, we need focus on just one of multicast and broadcast. In this paper, our explicit focus is broadcast. To broadcast from a source s, we use a broadcast tree T , which is a spanning tree of G that is rooted at s. The energy, e(u), required by a node of T to broadcast to its children is e(u) = max{w(u, v )|v is a child of u} Note that for a leaf node u, e(u) = 0. The energy, e(T ), required by the broadcast tree to broadcast a unit message from the source to all other nodes is e(T ) =
u

e(u)

For simplicity, we assume that only unit-length messages are broadcast. So, following a broadcast using the broadcast tree T , the residual energy, re(i, T ), at node i is re(i, T ) = ce(i) max{w(i, j )|j is a child of i in T } 0 The problem, MEBT, of nding a minimum-energy broadcast tree rooted at a node s is NP-hard, because MEBT is a generalization of the connected dominating set problem, which is known to be NPhard [20]. In fact, MEBT cannot be approximated in polynomial time within a factor (1 ), where is any small positive constant and is the maximal degree of a vertex, unless N P DT IM E [nO(log log n) ] [21]. Marks et al. [40] propose a genetic algorithm for MEBT and Das et al. [13] propose integer programming formulations. Wieselthier et al. [62, 63] propose four centralized greedy heuristicsDSA, MST, BIP, BIPPNto construct minimum energy broadcast trees. The DSA (Dijkstras shortest paths algorithm) heuristic (this is the BLU of [62, 63] augmented with the sweep pass of [62, 63]) constructs a shortest path from the source node s to every other vertex in G. The constructed shortest paths are superimposed to obtain a tree T rooted at s. Finally, a sweep is performed over the nodes of T . In this sweep, nodes are examined in ascending order of their index (i.e., in the order 1, 2, 3, , n). The transmission energy (i) for node 15

i is determined to be the max{w(i, j )|j is a child of i in T } If using (i) energy, node i is able to reach any descendents other than its children, then these descendents are promoted in the broadcast tree T and become additional children of i. The MST (minimum spanning tree) (this is the BLiMST heuristic of [62] augmented with a sweep pass) uses Prims algorithm [49] to construct a minimum-cost spanning tree (the cost of a spanning tree is the sum of its edge weights). The constructed spanning tree is restructured by performing a sweep over the nodes to reduce the total energy required by the tree. The BIP (broadcast incremental power) heuristic (this is the BIP heuristic of [62, 63] augmented with a sweep pass) begins with a tree T that comprises only the source node s. The remaining nodes are added to T one node at a time. The next node u to add to T is selected so that u is a neighbor of a node in T and e(T {u}) e(T ) is minimum. Once the broadcast tree is constructed, a sweep is done to restructure the tree so as to reduce the required energy. The BIPPN (broadcast incremental power per node) heuristic (this is the node-based MST heuristic of [63] augmented with a sweep pass) begins with a tree T that comprises only the source node s and uses several rounds to grow T into a broadcast tree. To describe the growth procedure, we dene e(T, v, i), where v T , to be the minimum incremental energy (i.e., energy above the level at which v must broadcast to reach its present children in T ) needed by node v so as to reach i of its neighbors that are not in T (of course, only neighbors j such that ce(v ) w(v, j ) are to be considered). Let R(T, v, i) = i/e(T, v, i). Note that R(T, v, i) is the inverse of the incremental energy needed per node added to T . In each round of BIPPN, we determine v and i such that R(T, v, i) is maximum. Then, to T , we add the i neighbors of v that are not in T and can be reached from v by incrementing v s broadcast energy by e(T, v, i). The i neighbors are added to T as children of v . Once the broadcast tree is constructed, a sweep is done to restructure the tree so as to reduce the required energy. Wan et al. [60] show that when w(i, j ) = c r d , the MST and BIP heuristics have constant approximation ratios and that the approximation ratio for DSA is at least n/2. Park and Sahni [46] describe two additional centralized greedy heuristicsMEN and BIPWLAto construct minimum energy broadcast trees. The second of these (BIPWLA) is an adaptation of the modied greedy algorithm proposed by Guha and Khuller [21] for the connected dominating set problem. The MEN (maximum energy node) heuristic attempts to use nodes that have more energy as non-leaf nodes of the broadcast tree (note that when i is a non-leaf, re(i) < ce(i), whereas when i is a leaf,

16

re(i) = ce(i)). In MEN, we start with T = {s}. At each step, we determine Q such that Q = {u|u is a leaf of T and u has a neighbor j , j T , for which w(u, j ) ce(u)} (1)

From Q, we select the node u that has maximum energy ce(u). All neighbors j of u not already in T and which satisfy w(u, j ) ce(u) are added to T as children of u. This process of adding nodes to T terminates when T contains all nodes of G (i.e., when T is a broadcast tree). Finally, a sweep is done to restructure the tree so as to reduce the required energy. The BIPWLA (broadcast incremental power with look ahead) heuristic is an adaptation of the look ahead heuristic proposed in [21] for the connected dominating set problem. This heuristic also may be viewed as an adaptation of BIPPN. In BIPWLA, we begin with a tree T that comprises the source node s together with all neighbors of s that are reachable from s using ce(s) energy. Initially, the source node s is colored black, all other nodes in T are gray and nodes not in T are white. Nodes not in T are added to T in rounds. In a round, one gray node will have its color changed to black and one or more white nodes will be added to T as gray nodes. It will always be the case that a node is gray i it is a leaf of T , it is black i it is in T but not a leaf, it is white i it is not in T . In each round, we select one of the gray nodes g in T ; color g black; and add to T all white neighbors of g that can be reached using ce(g) energy. The selection of g is done in the following manner. For each gray node u T , let nu be the number of white neighbors of u reachable from u by a broadcast the uses ce(u) energy. Let pu be the minimum energy needed to reach these nu nodes by a broadcast from u. Let, A(u) = {j |w(u, j ) ce(u) and j is a white node} We see that nu = |A(u)| and pu = max{w(u, j )|j A(u)}. For each j A(u), we dene the following analogous quantities B (j ) = {q |w(j, q ) ce(j ) and q is a white node} nj = |B (j )| pj = max{w(j, q )|q B (j )} Node g is selected to be the gray node u of T that maximizes nu /pu + max{nj /pj |j B (u)} Once the broadcast tree is constructed, a sweep is done to restructure the tree so as to reduce the required energy. 17

A localized distributed heuristic for MEBT that is based on relative neighborhood graphs has been proposed by Cartigny, Simplot and Stojmenovic [6]. Wu and Dai [64] develop an alternative localized distributed heuristic that uses k-hop neighborhood information. In a real application, the wireless network will be required to perform a sequence B = b1 , b2 , of broadcasts. Broadcast bi will specify a source node si and a message length li . Assume, for simplicity, that li = 1 for all i. For a given broadcast sequence B , the network lifetime is the largest i such that broadcasts b1 , b2 , , bi are successfully completed. The 6 heuristics of [62, 63, 46] may be used to maximize lifetime by performing each bi using the broadcast tree generated by the heuristic (the broadcast trees are generated in sequence using the residual node energies). However, since each of these heuristics is designed to minimize total energy consumed by a single broadcast, it is entirely possible that the very rst broadcast depletes the energy in a node, making subsequent broadcasts impossible. Singh, Raghavendra and Stepanek [51] propose a greedy heuristic for a broadcast sequence. The source node broadcasts to each of its neighbors resulting in a 2-level tree T with the source as root. T is expanded into a broadcast tree through a series of steps. In each step, a leaf of T is selected and its non-tree network neighbors added to T . The leaf selection is done using a greedy strategyselect the leaf for which the ratio (energy expended by this leaf so far)/(number of non-tree leaves) is minimum. The critical energy, CE (T ), following a broadcast that uses the broadcast tree T is dened to be CE (T ) = min{re(i, T )|1 i n} Park and Sahni [46] suggest the use of broadcast trees that maximize the critical energy following each broadcast. Let M CE (G, s) be the maximum critical energy following a broadcast from node s. For each node i of G, dene a(i) as below a(i) = {ce(i) w(i, j )|(i, j ) is an edge of G and ce(i) w(i, j )} Let l(i) denote the set of all possible values for re(i) following the broadcast. We see that l(i) = a(i) if i = s a(i) {ce(i)} otherwise

Consequently, the sorted list of all possible values for M CE (G, s) is given by L = sort(1in l(i)) We may determine whether G has a broadcast tree rooted at s such that CE (T ) q by performing either a breadth-rst or depth-rst search [49] starting at vertex s. This search is forbidden from using 18

edges (i, j ) for which ce(i) w(i, j ) < q . M CE (G, s) may be determined by performing a binary search over the values in L [46]. Each of the heuristics described in [62, 63, 46] to construct a minimum energy broadcast tree may be modied so as to construct a minimum energy broadcast tree T for which CE (T ) = M CE (G, s). For this modication, we rst compute M CE (G, s) and then run a pruned version of the desired heuristic. In this pruned version, the use of edges for which ce(i) w(i, j ) < M CE (G, s) is forbidden. Experiments reported in [46] indicate that this modication signicantly improves network lifetime, regardless of which of the 6 base heuristics is used. Lifetime improved, on average, by a low of 48.3% for the M EN heuristic to a high of 328.9% for the BIP P N heuristic. The BIP P N heuristic modied to use M CE (G, s), as above, results in the best lifetime.

3.3

Data Collection and Distribution

In the data collection problem, a base station is to collect sensed data from all deployed sensors. The data distribution problem is the inverse problem in which the base station has to send data to the deployed sensors (dierent sensors receive dierent data from the base station). In both of these problems, the objective is to complete the task in the smallest amount of time. Florens and McEliece [17, 18] have observed that the data collection and distribution problems are symmetric. Hence, once can derive an optimal data collection algorithm from an optimal data distribution algorithm and vice versa. Therefore, it is necessary to study just one of these two problems explicitly. In keeping with [17, 18], we focus on data distribution. Let S1 , , Sn be n sensors and let S0 represent the base station. Let pi be the number of data packets the base station has to send to sensor i, 1 i n. p = [p1 , p2 , , pn ] is the transmission vector. We assume that the distribution of these packets to the sensors is done in a synchronous time-slotted fashion. In each time slot, an Si may either receive or transmit (but not both) a packet. To facilitate the transmission of the packets, each Si has an antenna whose range is r . In the unidirectional antenna model, Si receives a packet only if that packet is sent in its direction from an antenna located at most r away. Because of interference, a transmission from Si to Sj is successful i the following are true: 1. j is in range, that is d(i, j ) r , where d(i, j ) is the distance between Si and Sj 2. j is not, itself, transmitting in this time slot 3. There is no interference from other transmissions in the direction of j . Formally, every Sk , k = i, that is transmitting in this time slot in the direction of Sj is out of range. Here, out of range means 19

d(k, j ) (1 + )r , where > 0 is an interference constant. In the omnidirectional antenna model, a packet transmitted by an Si is received by all Sj (regardless of direction) that are in the antennas range. The constraints on successful transmission are the same as those for the unidirectional antenna model except that all references to direction are dropped. Our objective is to develop an algorithm to complete the specied data distribution using the fewest number of time slots. A related data gathering problem for wireless sensor networks is considered in [71]. 3.3.1 Unidirectional Antenna Model

Florens and McEliece [17] develop an ecient algorithm to distribute data in a tree network using the fewest number of time slots. This algorithm is an extension of their data-distribution algorithm for a line network. We look only at the details of the line-network algorithm here. In a line network, S0 , , Sn are uniformly spaced on a straight line as in Figure 9. The base station, S0 is at the left end of the line and the spacing is g. We assume that (1 + )r/2 g r . Hence, when Si transmits a packet to its right, the packet can be received by Si+1 but not by Si+2 .

g S0 S1 S2 S3

Figure 9: A 3-sensor line network, S0 is the base station

Consider the transmission vector [2, 0, 0], which requires 2 packets be sent to S1 and 0 to the remaining sensors. This transmission may be completed in 2 time slots. S0 sends a packet in each of these two time slots to S1 . The transmission vector [0, 2, 0] requires 4 time slots. In the rst, S0 sends the rst packet to S1 . In the second time slot only one of S0 and S1 may transmit as if both transmit, then S1 will need to receive (from S0 ) and send (to S1 ) in the second time slot. This violates the restriction that a sensor cannot receive and send in the same time slot. To avoid buering problems at a sensor, we adopt a transmission discipline in which a packet is routed in consecutive slots until it reaches its destination. Adhering to this discipline requires that S1 transmit, in slot 2, the packet it received in slot 1. So, in slot 2, the base station S0 is idle. The remaining packet that is to be sent to S2 is sent in slots 3 and 4. Notice that [2, 0, 1] can be done is 4 slotsin slot 1 S0 sends to S1 , in slot 2 S1 sends to S2 , in slot 3 S0 sends to S1 and S2 sends to S3 , and in slot 4 S0 sends to S1 .

20

Step 1: [Transmit the packets for Sn , , S2 ] Transmit the packets for Sn , , S2 , in this order. For this transmission, use slots 2j 1, 1 j t, where t = n l=2 pl . The base station makes no transmission in slots 2j , 1 j t. Step 2: [Transmit S1 s packets] The packets destined for S1 are transmitted in slots 2t < j 2t + p1 . Figure 10: Base station algorithm of [17] for unidirectional antennas With the no buering discipline it is only the base station that has to make a decision for each time slot. If a sensor receives a packet that is destined for a sensor further down the line, it simply retransmits this packet in the next time slot. From our simple examples, it follows that the base station can transmit a packet in slot i + 1 i it either did not transmit a packet in slot i or the packet transmitted in slot i was destined for S1 . This feasibility constraint coupled with the greedy strategy to transmit packets in non-increasing order of destination distance results in the greedy base-station algorithm of Florens and McEliece [17] (Figure 10). Let L(p) be the last slot in which any of the Si s transmits a packet when the base station algorithm of Figure 10 is used. Florens and McEliece [17, 19] have shown that L(p) = max {i 1 + Pi + 2
1in i+1j n

pj }

and that this is the smallest possible value for L(p) over all feasible packet distribution strategies. Hence, the greedy algorithm of Figure 10 is an optimal distribution algorithm. The optimality of the greedy algorithm holds even if sensors are permitted to buer packets destined for other sensors. Notice that the transmission schedule for the data distribution task dened by p may be converted into a data collection schedule with the same L(p) value (for the data collection problem, pi is the number of packets that the base station is to collect from the base station from Si ). As noted in [17] each data distribution transmission from Si to Si+1 in time slot j becomes a data collection transmission from Si+1 to Si in time slot L(p) + 1 j . 3.3.2 Omnidirectional Antenna Model

The greedy unidirectional-antenna line-network algorithm of Figure 10 may be modied to obtain an optimal distribution algorithm for a line network that employs omnidirectional antennas (Florens and McEliece [18]). This modied line-network algorithm may then be extended to obtain an optimal distribution algorithm for trees [18]. Florens and McEliece [18] also propose a distribution algorithm for 21

Step 1: [Transmit the packets for Sn , , S3 ] Transmit the packets for Sn , , S3 , in this order. For this transmission, use slots 3j 2, 1 j t, where t = n l=3 pl . The base station makes no transmission in slots 3j and 3j 1, 1 j t. Step 2: [Transmit S2 s packets] The packets destined for S2 are transmitted in slots 3t + 2j 1, 1 j p2 . The base station makes no transmission in slots 3t + 2j , 1 j p2 . Step 3: [Transmit S1 s packets] The packets destined for S1 are transmitted in slots 3t + 2p2 < j 3t + 2p2 + p1 . Figure 11: Base station algorithm of [18] for omnidirectional antennas general networks. This distribution algorithm is within a factor 3 of optimal. As we did for the unidirectional antenna model, we consider only the case of line networks here. Consider the line network of Figure 9. We make the same assumptions as we did for the unidirectional modelsynchronous time-slotted transmissions, a sensor cannot transmit and receive a packet in the same slot, and sensors do not buer packets destined for other sensors. The unidirectional-antenna distribution strategy for p = [2, 0] and p = [0, 2] may be used for the omnidirectional-antenna case as well. However, the unidirectional strategy for p = [2, 0, 1] fails when applied to the case of omnidirectional antennas because of interference. In slot 3 of the unidirectional strategy, both S0 and S2 transmit a packet. Since S1 is within range of both S0 and S2 (when omnidirectional antennas are used), the interference at S1 causes the transmission from S0 to S1 to fail. In fact every successful distribution strategy for [2, 0, 1] must use at least 5 slots when the Si employ omnidirectional antennas. A possible 5-slot distribution strategy is to use the rst 3 slots to send S3 s packet (from S0 to S1 to S2 to S3 ) and then use slots 4 and 5 to send the 2 packets for S1 . With the no buering constraint, each sensor is required to forward, in the next slot, each packet it receives that is destined for another sensor. Only the base station has exibility in operation. So, we need only develop a base-station algorithm that determines the slot in which each packet (regardless of its destination) is transmitted to S1 . Because of the omnidirectional interference properties, we see that packets destined for S1 may be transmitted in successive slots, those destined for S2 may be transmitted in alternate slots, and those destined for Si , i > 2 may be transmitted only in every third slot. Coupling this observation with the greedy strategy of transmitting packets in non-increasing order of destination distance results in the greedy algorithm of Figure 11 [18]. The optimality of this algorithm is established in [18].

22

Sensor Fusion

The reliability of a sensor system is enhanced through the use of redundancy. That is, each point or region is monitored by multiple sensors. In a redundant sensor system, we are faced with the problem of fusing or combining the data reported by each of the sensors monitoring a specied point or region. Suppose that k > 1 sensors monitor point p. Let mi , 1 i k be the measurement recorded by sensor i for point p. These k measurements may dier because of inherent dierences in the k sensors, the relative location of a sensor with respect to p, as well as because one or more sensors is faulty. Let V be the real value for p. The objective of sensor fusion is to take the k measurements, some of which may be faulty, and determine either the correct measurement V or a range in which the correct measurement lies. The sensor fusion problem is closely related to the Byzantine agreement problem that has been extensively studied in the distributed computing literature [34, 14]. Brooks and Iyengar [3] have proposed a hybrid distributed sensor-fusion algorithm that is a combination of the optimal region algorithm of Marzullo [41] and the fast convergence algorithm proposed by Mahaney and Schneider [38] for the inexactagreement version of the Byzantine generals problem. Let i be the accuracy of sensor i. So, as far as sensor i is concerned, the real value at p is mi i (i.e., the value is in the range [mi i , mi + i ]). Each sensor needs to compute a range in which the true value lies as well as the expected true value. For this computation, each sensor sends its mi and i values to every other sensor. Suppose that sensor i is non-faulty. Then every non-faulty processor receives the correct values of mi and i . Faulty sensors may receive incorrect values. Similarly, if processor i is faulty, the remaining processors may receive diering values of mi and i . Figure 12 gives the Brooks-Iyengar hybrid algorithm [3]. This algorithm is executed by every sensor using as data the measurement ranges received from the remaining sensors monitoring point p plus the sensors own measurement. As an example computation, suppose that 4 sensors S 1, S 2, S 3 and S 4 monitor p and that the 4 measurement ranges are [2, 6], [3, 8], [4, 10] and [1, 7]. To perform the computation specied by the Brooks-Iyengar hybrid algorithm, the four sensors communicate their measurement ranges to one another. Assume that S 4 is the only faulty sensor. So, sensors S 1, S 2 and S 3 correctly communicate their measurement to one another. However, these sensors may receive diering readings from S 4. Likewise, S 4 may record dierent receptions from S 1, S 2 and S 3. Let the S 4 measurement received by S 1, S 2 and S 3 be [1, 3], [2, 7], [7, 12], respectively. The V and range computed at each of the 4 sensors are given in Figure 13. Note that k = 4 and = 1. 23

Step 1: [Determine range for real value V ] Let [li , ui , ni ], 1 i q be such that 1. li ui li+1 , 1 i < q and lq uq . The [li , ui ]s dene disjoint measurement intervals. 2. ni k gives the number of sensors whose measurement range includes [li , ui ]. 3. If x is a measurement value not included in one of the [li , ui ] intervals, x is included in the measurement interval of fewer than k sensors. V is estimated to lie in the range [l1 , uq ]. Step 2: [Estimate V ] V is estimated to be the weighted average
q i=1 [(li

+ ui ) ni ]/[2

q i=1 ni ].

Figure 12: Brooks-Iyengar hybrid algorithm to estimate V and range for V S1: The ranges recorded at S 1 are [2, 6], [3, 8], [4, 10] and [1, 3]. There is only one tuple [li , ui , ni ] that satises the Step 1 criteria. This tuple is [4, 6, 3]. S 1 estimates the range for V as [4, 6] and V = 5. S2: The recorded ranges are [2, 6], [3, 8], [4, 10] and [2, 7]. The tuples are [3, 4, 3], [4, 6, 4] and [6, 7, 3]. S 2 estimates the range for V as [3, 7] and V = 5. S3: The recorded ranges are [2, 6], [3, 8], [4, 10] and [7, 12]. The tuples are [4, 6, 3] and [7, 8, 3]. S 3 estimates the range for V as [4, 8] and V = 6.25. S4: Since this sensor is faulty, its computation is unreliable. It may compute any range and value for V. Figure 13: Example for Brooks-Iyengar hybrid algorithm

Conclusion

We have reviewed some of the recent advances made in the development of algorithms for wireless sensor networks. This paper has focussed on sensor deployment and coverage, routing (specically, unicast and multicast) and sensor fusion. Both centralized and distributed localized algorithms have been considered.

References
[1] J. Aslam, Q. Li and R. Rus, Three power-aware routing algorithms for sensor netowrk, Wireless Communications and Mobile Computing, 3, 2003, 187-208. [2] S. Banerjee and A. Misra, Energy ecient reliable communication for multi-hop wireless networks, WINET, 2001.

24

[3] R. Brooks and S. Iyengar, Robust distributed computing and sensing algorithm, IEEE Computer, June 1996, 53-60. [4] I. Caragiannis, C. Kaklamanis, and P. Kanellopoulos, New Results for energy-ecient broadcasting in wireless networks, Proc. of the 13th Annual International Symposium on Algorithms and Computation, 2002. [5] M. Cardei and D. Du, Improving wireless sensor network lifetime through power aware organization, ACM Wireless Networks, to appear. [6] Julien Cartigny, David Simplot, and Ivan Stojmenovic, Localized minimum-energy broadcasting in ad-hoc networks, IEEE INFOCOM, 2003. [7] K. Chakrabarty, S. Iyengar, H. Qi and E. Cho, Grid coverage for surveillance and target location in distributed sensor networks, IEEE Transactions on Computers, [8] J. Chang and L. Tassiulas, Routing for maximum system lifetime in wireless ad-hoc networks,, 37th Annual Allerton Conference on Communication, Control, and Computing, Monticello, IL, Sept. 1999. [9] J. Chang, and L. Tassiluas, Energy conserving routing in wireless ad-hoc networks, IEEE INFOCOM, 2000. [10] J. Chang and L. Tassiulas, Fast approximate algorithms for maximum lifetime routing in wireless ad-hoc networks, IFIP-TC6 Networking 2000, LNCS, 1815, Springer Verlag, 2000, pp. 702-713. [11] A.E.F Clementi, P. Crescenzi, P. Penna, G. Rossi, and P. Vocca, A worst case analysis of an MSTbased heuristic to construct energy-ecient broadcast trees in wireless networks, Technical Report 010 of the Univ. of Rome, 2001. [12] D. Culler and W. Hong, Wireless sensor networks, Special Issue, CACM, 47, 6, 2004. [13] Arindam K. Das, Robert J. Marks, and Mohamed El-Sharkawi, Minimum power broadcast trees for wireless networks, IEEE International Symposium on Circuits and Systems, May 2002. [14] D. Dolev, The Byzantine generals strike again, Jr. of Algorithms, 1982, 14-30. [15] S. Doshi, T. Brown, Minimum energy routing schemes for a wireless ad hoc network, IEEE INFOCOM, 2002. 25

[16] O. Egecioglu and T. Gonzalez, Minimum-energy broadcast in simple graph with limited node power, IASTED International Conference on Parallel and Distributed Computing and Systems, Anaheim, CA, August 2001, 334-338. [17] C. Florens and R. McEliece, Scheduling algorithms for wireless ad-hoc sensor networks, IEEE GLOBECOM, 2002, 6-10. [18] C. Florens and R. McEliece, Packets distribution algorithms for sensor networks, INFOCOM 2003. [19] C. Florens, M. Francesdhetti and R. McEliece, Lower bounds on data collection time in sensory networks, IEEE Jr. on Selected Areas in Communications, 1, 11, 2004. [20] M. Garey and D. Johnson, Computers and intractibility: A guide to the theory of NP-completeness, W. H. Freeman and Co., 1979. [21] Sudipto Guha and Samir Khuller, Approximation algorithms for connected dominating sets, Fourth Annual European Symposium on Algorithms, 1996. [22] W. Heinzelman, A. Chandrakasan and H. Balakrishnan, Energy-ecient communication protocol for wireless microsensor networks, IEEE HICSS, 2000. [23] A. Howard, M. Mataric and G. Sukhatme, An incremental self-deployment algorithm for mobile sensor networks, Autonomous Robots, Special Issue on Intelligent Embedded Systems. [24] A. Howard, M. Mataric and G. Sukhatme, Mobile sensor network deployment using potential elds: A distributed, scalable solution to the area coverage problem, 6th International Symposium on Distributed Autonomous Robotics Systems (DARS02), 2002. [25] C. Huang and Y. Tseng, The coverage problem in a wireless sensor network, WSNA, 2003. [26] S. Iyengar and R. Brooks, Computing and communications in distributed sensor networks, Special Issue, Jr. of Parallel and Distributed Computing, 64, 7, 2004. [27] S. Iyengar and R. Brooks, Handbook of Distributed Sensor Networks, Chapman & Hall/CRC, 2005. [28] R. Kannan, S.Sarangi, S.S. Iyengar and L. Ray, Sensor-centric quality of routing in sensor networks, INFOCOM, 2003. [29] R. Kannan, S. Sarangi, S. Ray and S. Iyengar, Minimal sensor integrity: Computing the vulnerability of sensor grids, Info. Proc. Letters, 86, 1, 2003, 49-55. 26

[30] R. Kannan and S. S. Iyengar, Game-theoretic models for reliable, path-length and energy-constrained routing in wireless sensor networks, IEEE Journal on Selected Areas in Communications, 2004. [31] R. Kannan, L. Ray, R. Kalidindi, and S. Iyengar, Energy threshold constrained geographic routing protocol for wireless sensor networks, Signal Processing Letters, 1, 1, 79-84, 2003. [32] K. Kar and S. Banerjee, Node placement for connected coverage in sensor networks, Proc WiOpt 2003: Modeling and Optimization in Mobile, Ad Hoc and Wireless Networks, 2003. [33] K. Kar, M. Kodialam, T. Lakshman and L. Tassiulas, Routing for network capacity maximization in energy-constrained ad-hoc networks, IEEE INFOCOM, 2003. [34] L. Lamport, R. Shostak, and M. Pease, The Byzantine generals problem. ACM Trans. Prog. Lang. Syst., July 1986, 499-516. [35] X. Li, P. Wan and O. Frieder, Coverage in wireless ad-hoc sensor networks, IEEE Transactions on Computers, 52, 2002, 753-763. [36] G. Lin and G. Noubir, Energy ecient broadcast with sectored antennas in wireless ad hoc networks, IASTED/IEEE Wireless and optical Communications, July 2002. [37] J. Lu and S. Tatsuya, Coverage-aware self-scheduling in sensor networks, IEEE Computer Communications Workshop (CCW 2003), 2003. [38] S. Mahaney and F. Schneider, Inexact agreement: Accuracy, precision, and graceful degradation, Fourth ACM Symp. on Principles of Distr. Computing, 1985, 237-249. [39] M. Maleki, K. Dantu, and M. Pedram, Power-aware source routing protocol for mobile ad hoc networks, SLPED02, 2002. [40] R. Marks, A. Das, M. El-Sharkawi, P. Arabshahi and A. Gray, Minimum power broadcast trees for wireless networks: Optimizing using the viability lemma, IEEE ISCAS, 2002. [41] K. Marzullo, Tolerating failures of continuous-valued sensors, ACM Trans. Computer Systems, Nov. 1990, 284-304. [42] S. Meguerdichian, F. Koushanfar, M. Potkonjak and M. Srivastava, Coverage problems in wireless ad-hoc sensor networks, IEEE InfoCom, 2001.

27

[43] S. Meguerdichian, F. Koushanfar, G. Qu and M. Potkonjak, Exposure in wireless ad hoc sensor networks, 7th Annual International Conference on Mobile Computing and Networking (MobiCom01), 2001, 139-150. [44] T. Melodia, D. Pompili, and I. Akyildiz, Optimal local topology knowledge for energy ecient geographical routing in sensor networks, IEEE INFOCOM, 2004. [45] A. Misra and S. Banerjee, MRPC: maximizing network lifetime for reliable routing in wireless,, IEEE Wireless Communications and Networking Conference (WCNC), 2002. [46] J. Park and S. Sahni, Maximum lifetime broadcasting in wireless networks, submitted. [47] S. Poduri and G. Sukhatme, Constrained coverage for mobile sensor networks, IEEE Intl. Conf. on Robotics and Automation (ICRA04), 2004, 165-171. [48] T. Rappaport, Wireless communications: Principles and practices, Prentice Hall, 1996. [49] S. Sahni, Data structures, algorithms, and applications in Java, 2nd Edition, Silicon Press, NJ, 2005. [50] S. Shakkottai, R. Srikant and N. Shro, Unreliable sensor grids: coverage, connectivity and diameter, INFOCOM, 2003. [51] S. Singh, C. Raghavendra, and J. Stepanek, Power-aware broadcasting in mobile ad hoc networks, IEEE PIMRC99, Osaka, Japan, Sep. 1999. [52] S. Singh, M. Woo, and C. Raghavendra, Power-aware routing in mobile ad hoc networks, ACM/IEEE MOBICOM, 1998. [53] S. Slijepcevic and M. Potkonjak, Power ecient optimization of wireless sensor networks, IEEE Intl. Conf. on Communications, 2001. [54] A. Spyropoulos and C. Raghavendra, Energy ecient communications in ad hoc networks using directional antenna, IEEE INFOCOM, 2002. [55] I. Stojmenovic, and Xu Lin, Power-aware localized routing in wireless networks, IEEE Transactions on Parallel and Distributed Systems, 2000. [56] R. Szewczyk, E. Osterweil, J. Polastre, M. Hamilton, A. Mainwaring and D. Estrin, Habitat monitoring with sensor networks, CACM, 47, 6, 2004, 34-40. 28

[57] D. Tian and N. Georganas, A coverage-preserving node scheduling scheme for large wireless sensor networks, [58] C. K. Toh, Maximum battery life routing to support ubiquitous mobile computing in wireless ad hoc networks, IEEE Communications Magazine, June, 2001, 138-147. [59] G. Veltri, Q. Huang, G. Qu and M. Potkonjak, Minimal and maximal exposure path algorithms for wireless embedded sensor networks, SenSys03, 2003. [60] P. Wan, G. Calinescu, X. Li and O. Frieder, Minimum-energy broadcast routing in static ad hoc wireless networks, IEEE INFOCOM, 2001. [61] X. Wang et al., Integrated coverage and connectivity conguration in wireless sensor networks, SenSys, 2003. [62] J. Wieselthier, G. Nguyen, and A. Ephremides, On the construction of energy-ecient broadcasting and multicast trees in wireless networks, IEEE INFOCOM, 2000. [63] J. Wieselthier, G. Nguyen, Algorithm for energy-ecient multicasting in static ad hoc wireless networks, Mobile Networks and Applications, 2001, 251-261. [64] J. Wu and F. Dai, Broadcasting in ad-hoc networks based on self pruning, IEEE INFOCOM, 2003. [65] J. Wu, M. Gao and I. Strojmenovic, On calculating power-aware connected dominating sets for ecient routing in ad hoc wireless networks, Jr. Communications and Networks, 4, 1, 2002. [66] X. Xu and S. Sahni, Optimal deployment of wireless sensors on a grid. [67] Y. Xu, J. Heidermann and D. Estrin, Geography-informed energy conservation for ad hoc routing, MOBICOM, 2001. [68] T. Yan, T. He, J. Stankovic, Dierentiated surveillance for sensor networks, First International Conference on Embedded Networked Sensor Systems, 2003, 51 - 62. [69] F. Ye, G. Zhong, S. Lu and L. Zhang, Peas: A robust energy conserving protocol for long-lived sensor networks, 23rd ICDCS, 2003. [70] F. Ye, G. Zhong, S. Lu and L. Zhang, Energy ecient robust sensing coverage in large sensor networks, Technical Report, UCLA, 2002. 29

[71] Y. Yu, B. Krishnamachari and V. Prasanna, Energy-latency tradeos for data gathering in wireless sensor networks, INFOCOM, 2004. [72] H. Zhang and J. Hou, Maintaining sensing coverage and connectivity in large sensor networks, Technical Report UIUC, UIUCDCS-R-2003-2351, 2003. [73] Y. Zou and K. Chakrabarty, Sensor deployment and target localization in distributed sensor networks, ACM Transactions on Embedded Computing Systems, 3, 1, 2004, 61-91.

30